An Unprecedented Confrontation Global NWP vs CALIPSO

Transcription

An Unprecedented Confrontation Global NWP vs CALIPSO
An Unprecedented Confrontation
Global NWP vs CALIPSO-CloudSat
10 Years of CALIIPSO-CloudSat
Paris, 8-10 June 2016
Richard Forbes
European Centre for Medium-Range
Weather Forecasts (ECMWF)
With thanks to Maike
Ahlgrimm, Alan Geer, Marta
Janisková, Frank Li, Katrin
Lonitz
Global Numerical Weather Prediction
! High resolution Earth System Modelling
Analysis and forecast system at ECMWF
• 
• 
• 
• 
• 
Global high resolution analysis and forecasts (dx=9 km) to 10 days
51 member ocn/atm coupled ensemble to 15 days (dx=18 km)
Ensemble of monthly and seasonal forecasts at lower resolution
COPERNICUS atmospheric composition analysis
Reanalyses ERA-I!ERA5 (1979-) & ERA-20C (20th Century)
Global prediction
•  Weather forecasts with focus on high impact weather
•  Longer term trends (monthly, seasonal)
•  Predicting predictability (representing uncertainty - ensembles)
Representing reality across space and time scales
•  NWP is an initial value problem, but not every aspect is constrained
•  The forecast quickly drifts towards its own “climate”
•  Need a physical and dynamical model as close to reality as possible
An accurate representation of clouds and precipitation is
vital!
Components
•  Atmosphere
•  Land,
hydrology
•  Ocean, sea-ice
•  Atmospheric
composition
The three relationship stages of global NWP vs. CALIPSO/CloudSat
1. Confrontation
(Model Evaluation)
2. ?
3. ?
Confronting models with observations
The first step is how to do a fair comparison
•  Both observational and modelling communities, we’ve been
learning!
•  Take the observations to the model (geophysical retrievals) or the model to the observations (forward modelling/
simulators)
•  We need both approaches
Examples:
•  CloudSat & CALIPSO level 2 and 3 products
•  DARDAR (Delanoë and Hogan, JGR, 2010)
•  COSP simulator package (CFMIP-Obs)
•  GOCCP (Chepfer et al 2010) A-train
GCM clouds differ significantly from CloudSat/CALIPSO
Differences are mainly due to deficiencies in model physics
Example: Vertical profile of cloud water content
Su et al. (2013)
Vertical profiles of cloud water content from CMIP5 models and CloudSat/CALIPSO as
fn(ω500)
GCM clouds differ significantly from CloudSat/CALIPSO
Differences are mainly due to deficiencies in model
physics Example:
Too frequent light precipitation over ocean
Stephens et al. (2010)
Fractional incidence (or frequency of occurrence) of rainfall as a function of
rainfall rate over ocean (60°S to 60°N) derived from CloudSat and various
global models.
GCM clouds differ significantly from CloudSat/CALIPSO
Differences are mainly due to deficiencies in model
physics
Example: Wide variation of supercooled liquid water fraction
McCoy et al. 2015
CMIP5 models and CALIPSO Obs
Cesana and Chepfer (2013), Cesana et al (2015), …
NWP analysis clouds differ significantly from CloudSat/
CALIPSO
Differences are mainly due to deficiencies in model physics
Millions of observations are assimilated every day at ECMWF …yet cloud ice/water content and phase are not well
constrained NWP analysis clouds differ significantly from CloudSat/
CALIPSO
Differences are mainly
duecontent
to deficiencies
Ice water
zonal cross- in model physics
sections
CloudSat
(non-convective,
non-precipitating
derived ice water)
(courtesy Frank Li)
mg
m-3
ERA-Interim
IFS with improved
cloud physics
operational in
2010
Spurious IWC max 0 to -23C due to
physics
assumptions
The three relationship stages of global NWP vs. CALIPSO/CloudSat
1. Confrontation
(Model Evaluation)
2.
Understanding
(Improving the model)
3. ?
Understanding processes - improving parametrizations
Using observation synergy and modelling studies
Example: Warm-rain formation process using A-Train data, GCMs and
process models.
Z vs Optical Depth for different Reff from
CloudSat/MODIS and from the HadGEM2 model
Suzuki et al.
2015
Effect of different
autoconversion
parametrizations
Understanding processes - improving parametrizations
Using observation synergy and modelling studies
Example: The Southern Ocean SW radiation bias
Forbes et al. 2016
(ECMWF Newsletter 146) Annual mean 10-20 Wm-2 TOA SW bias (too little reflection) over
Southern
ECMWF
lowOcean
resolution “climate”
ECMWF high resolution “analysis”
bias
dx=125 km 1 year forecast – CERES
bias
dx=16 km 24 hour forecast – CERES A snapshot of the IFS TOA SW radiation error shows the problem in
the IFS MODIS
visible
24 Aug 2013
IFS 24 hour
TOA net SW
radiation bias
vs CERES
24 Aug 2013
(red = not reflective
enough)
A snapshot of the IFS data assimilation system first guess departures
for SSMIS 37 GHz brightness temperatures (sensitive to LWP)
MODIS visible
24 Aug 2013
IFS SSMI/S
37GHz
analysis first
guess
brightness
temperature
errors
24 Aug 2013
(red = too little liquid
water)
The problem is a lack of supercooled liquid
water at the tops of convective clouds in cold-air
outbreaks
CALIPSO track
The problem is a lack of supercooled liquid
water at the tops of convective clouds in cold-air
outbreaks
CALIPSO
satellite lidar
cloud phase
IFS alongtrack lidar
forward
modelled
cloud phase
Parametrized convection and microphysics
Convection
Scheme
Convection
Scheme
Cloud
Scheme
Cloud
Scheme
Ice
processes
Ice
-23°
C
-23°
C
Super
cooled
Liquid
Mixe
d
0°C Liqui
d
0°C Liqui
d
Convection scheme microphysics:
Saturation adjustment,
autoconversion, detrainment to
cloud scheme
Phase = fn (T) from 0 to -23°C
Change to the model convective
parametrization to produce SLW at
colder temperatures
Supercooled liquid water now present at the tops
of convective clouds in cold-air outbreaks
CALIPSO
satellite lidar
cloud phase
IFS alongtrack lidar
forward
modelled
cloud phase
IFS with
convection
producing
SLW below
600hPa
More liquid water path (closer to SSMI/S) and SW
radiation dramatically reduced!
24 hour
forecast
Forbes et al. 2016
ECMWF Newsletter 146
Understanding processes - improving parametrizations
Using observation synergy and modelling studies
Example: The Southern Ocean SW radiation bias.
•  Synergy of observations from CERES, MODIS, SSMI/S &
CALIPSO helped to pinpoint the source of a major error in the
IFS
•  Improving the parametrization leads to improved forecasts
1 year low res “climate” forecast
The three relationship stages of global NWP vs. CALIPSO/CloudSat
1. Confrontation
(Model Evaluation)
2.
Understanding
(Improving the model)
3. Union
(Data assimilation)
A union of models and observations
Assimilation of active satellite data
•  If the model background state is far from the observations, it is very
difficult to extract useful information from the data.
•  Are the NWP systems good enough for active instruments (radar, lidar)
to have a positive impact? Does the data have a positive impact? •  ESA-funded projects at ECMWF over the last few years looking ahead
to the launch of EarthCARE and in readiness for assimilating the data
into the ECMWF global model.
From CALIPSO-CloudSat to EarthCARE
NASA Langley Research Center
ESA - AOES Medialab
Assimilating CALIPSO-CloudSat data
Janisková et al. (2012), Janisková (2014)
1D+4DVAR
Assimilating CALIPSO-CloudSat data
Janisková (2014)
1DVAR successfully draws model closer to the obs
•  1D-Var analysis of reflectivity, example track
•  Greatest impact in front, obs errors large in convection
Assimilating CALIPSO-CloudSat data
Janisková (2014)
1DVAR successfully draws model closer to the obs
+ve
impact
•  1DVAR brings model profile
closer to observed reflectivity,
lidar backscatter and optical
depth (independent)
•  Radar has largest impact
•  Lidar has impact near cloud
top
Assimilating CALIPSO-CloudSat data
Positive impact of 1D+4DVAR on the forecast
Zonal mean of rmse
difference for 24hr forecast
Blue is +ve
impact
Specific
Humidity
Temperatur
e
Janisková (2014)
Global mean of T+12,24,28
rmse difference
An Unprecedented Confrontation
Global NWP vs CALIPSO-CloudSat!
Summary
• 
Confrontation: CALIPSO & CloudSat have confronted GCMs
with an unprecedented level of cloud and precipitation
observations.
• 
Understanding: Multi-sensor observations have helped us to
understand processes and how to represent them in models.
• 
CALIPSO/CloudSat has inspired, and continues to inspire,
improvements to parametrizations.
• 
Union: Models getting closer to the observations. Assimilation of active sensors becoming feasible & beneficial
• 
10 years so far and the legacy of CALIPSO/CloudSat for NWP
and climate model development continues with vigour…